Matches in SemOpenAlex for { <https://semopenalex.org/work/W4313018220> ?p ?o ?g. }
Showing items 1 to 99 of
99
with 100 items per page.
- W4313018220 endingPage "708" @default.
- W4313018220 startingPage "691" @default.
- W4313018220 abstract "Most state-of-the-art deep neural networks use static inference graphs, which makes it impossible for such networks to dynamically adjust the depth or width of the network according to the complexity of the input data. Different from these static models, depth-adaptive neural networks, e.g. the multi-exit networks, aim at improving the computation efficiency by conducting adaptive inference conditioned on the input. To achieve adaptive inference, multiple output exits are attached at different depths of the multi-exit networks. Unfortunately, these exits usually interfere with each other in the training stage. The interference would reduce performance of the models and cause negative influences on the convergence speed. To address this problem, we investigate the gradient conflict of these multi-exit networks, and propose a novel meta-learning based training paradigm namely Meta-GF (meta gradient fusion) to harmoniously train these exits. Different from existing approaches, Meta-GF takes account of the importances of the shared parameters to each exit, and fuses the gradients of each exit by the meta-learned weights. Experimental results on CIFAR and ImageNet verify the effectiveness of the proposed method. Furthermore, the proposed Meta-GF requires no modification on the network structures and can be directly combined with previous training techniques. The code is available at https://github.com/SYVAE/MetaGF ." @default.
- W4313018220 created "2023-01-05" @default.
- W4313018220 creator A5052883326 @default.
- W4313018220 creator A5064240819 @default.
- W4313018220 creator A5078737459 @default.
- W4313018220 date "2022-01-01" @default.
- W4313018220 modified "2023-10-05" @default.
- W4313018220 title "Meta-GF: Training Dynamic-Depth Neural Networks Harmoniously" @default.
- W4313018220 cites W2194775991 @default.
- W4313018220 cites W2331143823 @default.
- W4313018220 cites W2884751099 @default.
- W4313018220 cites W2895387432 @default.
- W4313018220 cites W2962944050 @default.
- W4313018220 cites W2963163009 @default.
- W4313018220 cites W2963393494 @default.
- W4313018220 cites W2963791342 @default.
- W4313018220 cites W2981812042 @default.
- W4313018220 cites W2981884310 @default.
- W4313018220 cites W2996151060 @default.
- W4313018220 cites W3024979215 @default.
- W4313018220 cites W3034292689 @default.
- W4313018220 cites W3034421924 @default.
- W4313018220 cites W3035038672 @default.
- W4313018220 cites W3035071066 @default.
- W4313018220 cites W3035424951 @default.
- W4313018220 cites W3037913581 @default.
- W4313018220 cites W3081980877 @default.
- W4313018220 cites W3090786842 @default.
- W4313018220 cites W3094307698 @default.
- W4313018220 cites W3096609285 @default.
- W4313018220 cites W3107016329 @default.
- W4313018220 cites W3109269658 @default.
- W4313018220 cites W3112673818 @default.
- W4313018220 cites W3128411221 @default.
- W4313018220 cites W3176694147 @default.
- W4313018220 cites W3181848549 @default.
- W4313018220 cites W3204647170 @default.
- W4313018220 cites W4214700987 @default.
- W4313018220 doi "https://doi.org/10.1007/978-3-031-20083-0_41" @default.
- W4313018220 hasPublicationYear "2022" @default.
- W4313018220 type Work @default.
- W4313018220 citedByCount "2" @default.
- W4313018220 countsByYear W43130182202023 @default.
- W4313018220 crossrefType "book-chapter" @default.
- W4313018220 hasAuthorship W4313018220A5052883326 @default.
- W4313018220 hasAuthorship W4313018220A5064240819 @default.
- W4313018220 hasAuthorship W4313018220A5078737459 @default.
- W4313018220 hasConcept C11413529 @default.
- W4313018220 hasConcept C119857082 @default.
- W4313018220 hasConcept C153258448 @default.
- W4313018220 hasConcept C154945302 @default.
- W4313018220 hasConcept C162324750 @default.
- W4313018220 hasConcept C177264268 @default.
- W4313018220 hasConcept C187736073 @default.
- W4313018220 hasConcept C199360897 @default.
- W4313018220 hasConcept C2776214188 @default.
- W4313018220 hasConcept C2776760102 @default.
- W4313018220 hasConcept C2777303404 @default.
- W4313018220 hasConcept C2780451532 @default.
- W4313018220 hasConcept C2781002164 @default.
- W4313018220 hasConcept C41008148 @default.
- W4313018220 hasConcept C45374587 @default.
- W4313018220 hasConcept C50522688 @default.
- W4313018220 hasConcept C50644808 @default.
- W4313018220 hasConceptScore W4313018220C11413529 @default.
- W4313018220 hasConceptScore W4313018220C119857082 @default.
- W4313018220 hasConceptScore W4313018220C153258448 @default.
- W4313018220 hasConceptScore W4313018220C154945302 @default.
- W4313018220 hasConceptScore W4313018220C162324750 @default.
- W4313018220 hasConceptScore W4313018220C177264268 @default.
- W4313018220 hasConceptScore W4313018220C187736073 @default.
- W4313018220 hasConceptScore W4313018220C199360897 @default.
- W4313018220 hasConceptScore W4313018220C2776214188 @default.
- W4313018220 hasConceptScore W4313018220C2776760102 @default.
- W4313018220 hasConceptScore W4313018220C2777303404 @default.
- W4313018220 hasConceptScore W4313018220C2780451532 @default.
- W4313018220 hasConceptScore W4313018220C2781002164 @default.
- W4313018220 hasConceptScore W4313018220C41008148 @default.
- W4313018220 hasConceptScore W4313018220C45374587 @default.
- W4313018220 hasConceptScore W4313018220C50522688 @default.
- W4313018220 hasConceptScore W4313018220C50644808 @default.
- W4313018220 hasLocation W43130182201 @default.
- W4313018220 hasOpenAccess W4313018220 @default.
- W4313018220 hasPrimaryLocation W43130182201 @default.
- W4313018220 hasRelatedWork W1486452452 @default.
- W4313018220 hasRelatedWork W1980008589 @default.
- W4313018220 hasRelatedWork W2082482750 @default.
- W4313018220 hasRelatedWork W2544122622 @default.
- W4313018220 hasRelatedWork W2961085424 @default.
- W4313018220 hasRelatedWork W3199608561 @default.
- W4313018220 hasRelatedWork W4285170134 @default.
- W4313018220 hasRelatedWork W4306674287 @default.
- W4313018220 hasRelatedWork W4319309271 @default.
- W4313018220 hasRelatedWork W4224009465 @default.
- W4313018220 isParatext "false" @default.
- W4313018220 isRetracted "false" @default.
- W4313018220 workType "book-chapter" @default.